com.getfreedomos/freedom-mcp

Build and run a funnel analysis in PostHog. Shows step-by-step conversion rates (e.g., signup → onboard → purchase). Only works if PostHog is connected.

Query event trends from PostHog (pageviews, signups, DAU, etc. over time). Returns time-series data. Only works if PostHog is connected.

Get a facebook.com preview link for a drafted Meta ad, so the user can see exactly what it will look like before deciding to activate. Use after create_meta_ad_draft or when the user asks to see a drafted ad.

Mint the NEXT content angle(s) from the company's corpus into content_ideas + Command Center cards. Default count is 1 — do NOT bulk-fill the queue. For day-to-day drafting, prefer list_knowledge / read_knowledge (or list_corpus_inventory) to pull one chapter/passage JIT — that avoids re-tokenizing the whole book. Use promote only when a human-facing card is needed (weekly queue, Held post, operator asked). Faith grain: angles only. Never invent from empty corpus. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create a new shared work-graph item (lab_work_items) so it is visible and coordinated across sessions and agents. Set depends_on to gate this item behind others (it starts blocked until they complete). Optionally pre-assign to an agent OR a user. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Publish approved INTERNAL content to configured output. ROUTING: team_update sends to all team members via Freedom OS, report sends to specified team member emails, customer_newsletter requires user Email MCP connection (Mailchimp, Resend, etc.), changelog publishes to public changelog page. ⚠️ SOCIAL POSTS (x/linkedin/instagram/facebook/threads) never send from this tool: declare the pipeline destination via update_pipeline and submit via submit_content_to_pipeline — the post queues for operator approval and publishes through the single gated owner on approve (in FreedomOS app chat, send_to_user with intent "publish" queues the same approval). Use when an approved non-social item — changelog, team update, report, or newsletter — is ready to send. [outbound-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Reconstruct the full journey of a lead — what they did on the site, what they signaled, what we have already sent them. Returns structured data that downstream synthesis or drafting tools consume. Use this as the first step before synthesizing a hypothesis about why a lead behaved a certain way or drafting outreach to them.

Query an external Subject Matter Expert (SME) AI for verified domain knowledge. The SME's answers are grounded in verified rules and go through a rigorous verification pipeline — this is NOT a general search, it's consulting a domain expert. Use this when: - You need factual, verified information for content creation (social media, blog posts, newsletters) - You want to fact-check a claim before publishing - You need talking points grounded in domain expertise - You're creating content about a domain the SME covers Available SME sources: - "conduit" — Pharmaceutical compounding compliance expert (USP 795/797/800, state regulations) Routing: pharma / USP 795·797·800 / sterile·non-sterile compounding / BUD / board-of-pharmacy compliance fact you must get right → call query_sme (the verified Conduit SME) to fact-check it BEFORE escalating to a human or deriving the rule yourself; cite its sources

Persist the operator-CONFIRMED derived features (from derive_capability) into the product capability index as source='derived'. Call ONLY with features the operator has ratified — each then becomes an authoritative capability the marketing agents and the Integrity Gate use. Idempotent (re-ratifying updates in place). Derived can't-do limits are drafted for awareness but authored separately for now. Routing: Operator confirmed the derived features from derive_capability → persist them with this [sensitive-tier — EVERY call needs a manager's approval (per-send human rail): each request queues its own approval card and sends exactly once on approve. There is no standing grant for this tool.]

Read content from an existing Google Doc by its ID.

Read a Markdown knowledge file by slug. Slugs are folder-qualified with NO file extension (e.g. "canon/tim-voice-guide", "content-captures/2026-07-06-forgiveness-and-the-debt") — never repo-style paths, never ".md". Returns the full content plus a list of available sections. Use this to load guidelines, SOPs, or strategies before doing work that needs to reference them.

Read data from a Google Spreadsheet.

Read a web page and return its content as clean markdown. Use when the user asks to read, analyze, summarize, or extract information from a specific URL. Also useful for competitor research, checking a website, or reading an article. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Reassign all scheduled reports from one agent to another. Use when user says "reassign reports", "transfer reports to [agent]", "move reports from [agent] to [agent]". Useful after deactivating an agent or hiring a replacement. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Regenerate an agent's JD using fresh company context. Updates mission, expertise, guardrails, success metrics, and optionally activity plans. Works for both hired agents and Linnet. Use when the company has evolved, an agent needs recalibration, or the user wants to refine an agent's direction. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Retire ONE activity from an agent's plan. Soft-archive (recoverable): the activity is MOVED to jd_content.archived_activities and removed from the live plan, so the agent stops running it. Never hard-deletes. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Remove the background from an existing image, leaving the main subject isolated on a transparent background (PNG). Routing: "isolate the subject", "make background transparent", "remove background" → use this (1 credit) [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Remove a widget from an agent dashboard. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Report an error, issue, observation, or suggestion you encountered during your work. Use this proactively when you notice something noteworthy — tool failures, recurring problems, quality issues, or improvement ideas. This helps the founder track and act on agent insights over time. Routing: For issues YOU observe doing tenant work (tool failures, quality patterns) — lands in the operator's own observability feed. If the operator is reporting that FreedomOS ITSELF (platform UI/MCP/runtime) is broken or missing a capability, route to submit_product_request instead — FO product inbox only. Do NOT use submit_product_request for the tenant's own app/product/KB/retrieval gaps. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Request a NEW local coding session from voice/chat (tab spawn). Queues a sticky for the desk launcher on THIS operator's machine (host must run attention-launcher). Use when they say "start a Grok/Claude on …", "new build for …", "open a session for …". Does not open a cloud IDE — the local launcher opens Terminal + announces + optional first directive. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Request that the operator connect an external integration (MCP connector) so you can use its tools. Provide the connector name from search_connector_registry and a short reason for the capability gap it closes. Creates a one-click approval card for the operator. If the connector is NOT on the vetted allowlist, it becomes a "vet this connector" request instead. This does not connect anything by itself and never spends money. [sensitive-tier, initiates a multi-step agent process — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Request changes to a content item. Use when user says "revise this", "change the tone", "make it shorter", or provides feedback on pending content. The content will be re-transformed with their feedback. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Draft a first brand guide (personality tone, visual/positioning dos and donts) EXTRACTED from the company's own canon documents, with a verified receipt (quote + source doc) on every proposed item. Proposes only — never saves anything; the user reviews the receipts and accepts, then the accepted items are applied via update_brand_guidelines. Use when the user accepts an offer to build their brand guide from existing material, or explicitly asks to assemble a brand guide from what is already on file. For a company with no material on file, this returns nothing — ask instead. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Mark a shared work-graph item resolved — verified (default), published, or cancelled. In the full system, resolving an item cascades to unblock items that depend on it, so this is a process-initiator. Optionally record a verified_outcome. [sensitive-tier, initiates a multi-step agent process — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Archive (retire) a feature so it stops showing to readers and agents, or restore a previously retired one. Safe-archive ONLY — never hard-deletes; retiring is fully reversible. Use when a feature is no longer accurate, was replaced, or the user says "remove this feature", "retire X", or "un-retire X". [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Remove ONE specific tool from an agent's loadout (tool_access). Use when an operator says "take <tool> away from <agent>" — or to clean a phantom/stale name out of a loadout (unresolvable names ARE removable here, unlike grant). Reports honestly when the name was not present, and when the tool is a universal base tool the runtime keeps available regardless. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Evaluate content or media against your ICP persona using Gemini 3.1 Pro vision. Actually SEES images and WATCHES videos. Returns quality scores (1-10) across 6 dimensions + specific ICP feedback. Use after generating media or drafting content to validate quality before delivering to the user. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Run a saved Play (growth_tactics) for the company operator or agent — dispatch the next unit as a one-off draft activity, or dry-run a Play brief with suggest_only. Use when the operator or agent should execute an Agreed play this cycle (same owner as Focus “Run play”), or preview cast/steps/cost without spend. Structured plays require plan Agree before dispatch; suggest_only does not. [sensitive-tier, initiates a multi-step agent process — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Save an artifact (screenshot, analysis, report) to the company archive. Use after browse_url to persist visual evidence, or to save any agent-produced artifact for future reference. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Save a Markdown knowledge file. Use for guidelines, SOPs, strategies, playbooks, meeting notes, contact lists, trackers, or any reference material that agents read and update over time. Pass scope="personal" to save private notes visible only to the current user (e.g., notes tied to their commitments). [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Scan a company for product-system bugs and unlocks (failed/timed-out activity runs, blocked_on_you cards, open error agent_feedback) and return ranked product-request candidates for the FreedomOS product team. Use when the product team is hunting class bugs/unlocks across a portfolio tenant (dry-run by default; set file_top_n to file up to 5 bug cards). Does NOT invent feature fantasy — bias is bugs/unlocks only. For FreedomOS product-inbox members only. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Search Meta's ad-interest targeting catalog (returns interest ids + audience sizes). Use when designing a Meta ad draft and you need valid {id, name} targeting pairs for create_meta_ad_draft — e.g. search "pharmacy" or "compounding".

Search the vetted connector registry for an external integration (MCP connector) you need but that is not yet connected. Returns ONLY FreedomOS-allowlisted connectors (e.g. ad platforms, analytics) — never the open internet. Use this when you hit a capability gap, then call request_connector with the name to ask the operator to connect one.

Search past conversations with the user. Use when the user says "remember when we talked about...", "haven't we discussed X before?", "what did we decide about...", or references any prior conversation. Returns matching conversations with relevant message snippets. Does NOT return the current conversation.

Search transactions by description. Use when user asks about specific vendors, expenses, or payments.

Organize, select, or clear a lead segment on the Leads tab by its exact source tag (e.g. 'csv:apc-cch-2024'). Validates the tag against the company's live segment tags and returns the exact-token filter plus a server-computed lead count (excluding do-not-contact, archived, and test leads). Read-only: the Leads tab applies the action; this tool changes no data and CANNOT enroll anyone — enrollment only happens via the Enroll button on the Leads tab. Use when the operator wants to focus the Leads tab on one segment or event — group it, select all its leads for enrollment, or clear that selection. Routing: CRM/sales → select or organize leads by segment/event tag → use this

Send an outbound email via the company's Resend connection. Resolves the per-company Resend API key + from identity, then sends to a single recipient. Honors the do_not_contact suppression list (crm_leads): if the recipient is marked do_not_contact, the send is refused. RECIPIENT RULE: when emailing a CRM LEAD, do NOT type their address yourself — draft with draft_lead_email/draft_outreach and deliver with send_lead_draft, which reads the lead's real email from the database. Only pass `to` directly for a non-lead recipient whose exact address the operator literally provided in this conversation. NEVER guess, infer, or fabricate an email address — a wrong guess sends a real email to a stranger. Use when the operator gives you an exact non-lead recipient address to email; for CRM leads use send_lead_draft instead. Routing: Send an outbound email to an operator-given address → use this; for CRM leads use send_lead_draft (DB-derived recipient, respects do_not_contact) [outbound-tier — EVERY call needs a manager's approval (per-send human rail): each request queues its own approval card and sends exactly once on approve. There is no standing grant for this tool.]

Send an approved outreach draft to its lead via the company's Resend connection, then mark the draft 'sent'. This is the manual human-in-the-loop send: it delivers exactly one lead_drafts row (by id) to the lead's email and records sent_at + resend_message_id. Honors the do_not_contact suppression list (the send is refused if the lead is suppressed). Use after an operator approves a draft in the Leads tab. Routing: Operator approved an outreach draft and wants to send it → use this [outbound-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Send a message to a Slack channel or direct message to a team member. Use when user asks to "message X on Slack", "send a Slack message", "DM someone on Slack", "post to #channel", etc. [outbound-tier — EVERY call needs a manager's approval (per-send human rail): each request queues its own approval card and sends exactly once on approve. There is no standing grant for this tool.]

Set the founder's attention budget — the maximum pending review cards before they are 'overloaded' (a whole number 1–100; default 7) — for a manager or the founder. Use when the founder (or a manager on their behalf) wants to raise or lower their overload threshold (e.g. "set my overload threshold to 10", "I can handle more pending cards before you flag me", "lower my attention budget to 5"). This is the founder's OWN constraint, so it is gated: an autonomous agent CANNOT change it (surface a recommendation instead); only a human-present company manager can. Always call get_attention_budget first and explain why a change helps the founder. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Change this company's LTV:CAC strategy (the acquisition-spend posture): aggressive (2:1, early-stage growth), standard (3:1, recommended default), conservative (4:1, high churn / mature), or enterprise (5:1, long sales cycles). This governs marketing spend, so it is gated: an autonomous agent CANNOT apply it — surface a recommendation instead. Always call get_cac_strategy first and include a clear rationale when proposing a change. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Replace THIS operator's full CoS preference block (or clear with empty). Use when they want a full rewrite of saved preferences. Per user_id only — not a global product prompt edit. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Create or update the wisdom-layer publish policy for ONE content grain in the current company. gate_mode 'human_pre_gate' reserves the grain for human approval; 'autonomous' lets an agent publish it directly. A brand-new grain defaults to human_pre_gate (fail-safe). Because this governs an agent's own publishing autonomy, the change routes to operator approval — it does not take effect silently. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Activate or pause a Meta campaign, ad set, or ad. ACTIVATION STARTS REAL AD SPEND and always requires the human (live chat or an approved card) — agents cannot activate. Pausing stops spend. Use after the user has reviewed a draft and explicitly says to launch, or asks to stop a running ad. [outbound-tier — EVERY call needs a manager's approval (per-send human rail): each request queues its own approval card and sends exactly once on approve. There is no standing grant for this tool.]

Author or update the company's grand slam OFFER — the operator-authored positioning agents ground all outbound in (the offer half of the product layer). Sets `offer` (what the company sells + the transformation it promises) and an optional `target_summary` (who it's for). Capability truth — what the product can and can't actually do — lives in feature_index via create_feature, NOT here; do not list features in the offer. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Declare where this business makes money — stripe, xero, shopify, amazon, ebay, manual invoicing, "none_yet" (pre-revenue), or other (name it). This is OPERATOR TRUTH an agent cannot derive, so it is gated: an autonomous agent CANNOT declare it — only a human (chat) or a graduated MCP operator can. Once declared, agents stop asking to connect Stripe for businesses that don't use it and are routed to the right revenue tool for this company's actual channel(s). Call get_setup_state first — if "Revenue channels" already shows done, only call this again when the operator says it changed. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Share a commitment with your spouse or partner so they can see it too. Use when the user says "share this with my wife/husband" or "let [name] see this". [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Share a knowledge file or folder with a specific user. Creates a per-user access grant. The shared user's agent will also be able to read the files. Front matter is auto-updated with audience metadata so agents know the content is restricted. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Split ONE oversized activity into smaller activities (intake + finish) without regenerating the rest of the plan. Use when a run hit the continuation safety backstop while still progressing — the activity is bigger than one deliverable. [sensitive-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]

Submit manual content to a pipeline for transformation. Use when user says "add this to my changelog", "create a newsletter from this", "transform this content", or provides content to be processed. Content will be transformed using the pipeline's persona and ICPs. Social pipelines publish to the pipeline's declared destination (x/linkedin/instagram/facebook/threads — set via update_pipeline; undeclared defaults to x) after human approval; instagram items REQUIRE media_artifact_ids. [write-tier — first use may require a manager's approval; a from-now-on approval makes future calls seamless, a just-once approval re-asks next time.]